In a previous paper, meteorological circumstances of myocardial infarctions, cerebrovascular attacks, and suicidal attempts were studied by a univariate method. The present work used the same clinical reports, collected by the Medical Emergency Assistance System (SAMU) in the Paris area from 1975 to 1977, but with multivariate calculations. 150 potential predictive indicators were submitted to "progressive ascending selection". Selected indicators were then combined into a composite index by "linear canonic discrimination". This index was tested in terms of successful prediction. The 150 indicators were: 1) meteorological variables, recorded at ground level, such as wind and temperature (expressed respectively in 28 and 24 ways), airpressure, moisture; 2) variables computed from data recorded in altitude; 3) pollutants; 4) non-meteorological indicators, such as day of the week, season, solar activity; 5) the "past of the predictand", i.e. the frequency of infarctions during the previous days; 6) types of weather, defined after confronting meteorological maps with clinical data. The coding of qualitative data required a new procedure. The event to be predicted, which occurred only one day a week, was an incidence of infarctions of at least twice the average. The percentage of successful prediction was 78.7%. The type of weather was by far the best indicator. Detrimental circumstances were changing weathers, with in the order of decreasing correlations, atmosphere fluxes coming from S-SE, E, SW, and NW. These results complete those of univariate analysis. They validate a simple and efficient predictive method, similar in its principle to that used in Germany.